AI and Sustainability Workshop

The Data Center Expansion

In the context of the rapid expansion of data centers in France and across Europe, the Observatory on the Environmental Footprint of AI is bringing together researchers, experts, and policymakers to discuss the environmental challenges associated with the development of artificial intelligence infrastructure.

MORNING: GENERAL PUBLIC

9h30 – 9h50: Coffee

09h50 – 10h00: Welcome remarks

École normale supérieure

AI and Society Institute

Observatory on the Environmental Footprint of AI

10h00 – 10h40: Keynote (40 min)

(forthcoming)

Cecilia Rikap: author of The Rulers: Corporate Power in the Age of AI and the Cloud

10h40 – 11h20: Are all AI use cases worth it?

15 min - Tristan Coignon: The usage and environmental impacts of AI video

generation

Abstract: In the past years, AI video generation has gone from a technology barely able to produce anything meaningful to a fully-fledged tool used by millions across the world. However, generating AI video comes at a steep resource cost and the broader environmental impacts of AI video generation as a technology, and their evolution over time, are still understudied. As such, in this presentation, we aim to model and estimate the conservative worldwide impacts of AI video generation over the years.

15 min - Antonia Marcheva: Costs and Benefits of Generative AI in the Global

Economy

Abstract: The discussion of the impact of generative AI is divided into groups who believe it will primarily improve lives, and those who believe the environmental and social consequences will overwhelm any positive impact. We use a state-of-the-art model of the global economy to explore the tension between the environmental costs, the increase in useful economic goods (e.g. healthcare and education), and the change in the structure of the economy (e.g. jobs). For environmental costs, we focus on the increase in emissions from the electricity sector, both from direct electricity use and the failure to phase out the most polluting electricity generation due to increased demand. We also pay particular attention to which countries receive the benefits of generative AI and which ones bear the costs.

10 min: Q&A

11h20 – 11h40: Break

11h40 – 12h15: Panel TBD - Anticipating token consumption

Retrieval pipelines, reasoning models and agentic workflows multiply the number of tokens behind a single user request. Organisations deploying generative AI can thus underestimate how many tokens (and energy) they will actually consume. This panel brings together organisations that have had to put their token consumption under management: how they forecast it, where the overruns came from, and what they changed in response.

12h15 – 12h55: Challenges of the data center expansion

15 min – Loup Cellard: The many barriers to waste heat reuse from data centres.

Abstract**: In order to improve the urban development of data centres, operators are** trying to give something back to local municipalities. Data centres can then supply heat to a closed local loop or an urban heating network. How can we explain the difficulty in matching the heat supply from data centres with the demand for decarbonised energy in local areas? The talk will present the results of an empirical study conducted in France since January 2026.

15 min – Clément Gasull: Grid connection regimes for data centers in France

10 min - Q&A

12h55 – 13h: Closing remarks